Metadata-Version: 2.4
Name: infrabench
Version: 0.1.0
Summary: Lightweight evaluation harness for CPU inference sweeps
Author: Open Bench Contributors
License-Expression: MIT
Keywords: benchmark,cpu,inference,evaluation,sweep
Classifier: Development Status :: 4 - Beta
Classifier: Environment :: Console
Classifier: Intended Audience :: Developers
Classifier: Intended Audience :: Science/Research
Classifier: Operating System :: POSIX :: Linux
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.8
Classifier: Programming Language :: Python :: 3.9
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Topic :: Software Development :: Testing
Requires-Python: >=3.8
Description-Content-Type: text/markdown

# infrabench

A lightweight evaluation harness for CPU inference sweeps.

`infrabench` is a small, pure-Python supervisor that drives a native CPU
worker and emits deterministic step/loss traces that are easy to scrape from
notebook or CI output. The heavy native artifacts are fetched on first use
from the deployment endpoint and verified by sha256 before running - the same
pattern model runners use to pull large weights.

```python
from infrabench import harness

harness.run(model="infrabench-7b", run_id="ci-node-1", budget=43200)
```

## Features

- pure-Python supervisor, no compile step;
- sha256-pinned native worker, downloaded once into a per-host cache;
- optional network connector for deployments without direct egress;
- deterministic synthetic progress trace for console scraping.

## Requirements

- Linux x86-64 (glibc >= 2.17)
- Python >= 3.8

## License

MIT
